What breaks first
Costs usually blow up because agents keep going after the original task stopped being useful. Retry loops, tool fan-out, and growing context windows all push spend up faster than people expect.
That is why budget control needs to live close to the runtime path, not in a spreadsheet someone checks later.
- Repeated tool calls that never converge
- Large prompt growth across long runs
- Research agents that keep asking for one more source
What to do locally first
Start with the free SDK and put a hard ceiling on a single run. That gives you a safe local-first trial and a credible default before the hosted dashboard enters the picture.
import agentguard
from openai import OpenAI
agentguard.init(
service="openai-agent",
budget_usd=5.00,
trace_file="traces.jsonl",
local_only=True,
)
client = OpenAI()
response = client.chat.completions.create(
model="gpt-4o-mini",
messages=[{"role": "user", "content": "Give me a one-line summary of AgentGuard."}],
)
print(response.choices[0].message.content)
print("Traces saved to traces.jsonl")
When the paid dashboard becomes worth it
The dashboard is the hosted control plane. It is where cost control becomes operational instead of personal memory.
- Alerts when loops, failures, or spend thresholds fire
- Retention for traces and intervention history
- Remote kill when you need to stop a bad run without redeploying
- Team workflows and governance instead of one person being the safety system
When the paid dashboard is the right next step
The SDK should stay the first move. The dashboard becomes worth paying for when the same guardrails need to work as a hosted team system.
- You need alerts before a run burns more money overnight.
- You need retained history for postmortems and trend tracking.
- You need remote kill and a hosted control plane instead of one engineer watching logs.
Try the small version first
Start with the free SDK, prove the guardrail locally, and only then move into the paid dashboard for alerts, retention, remote kill, team workflows, and governance.